One of the most exciting developments for contact centers over the last several years has been the increasing sophistication of analytics software, allowing for the analysis of up to 100% of customer calls. The Natural Language Processing (NLP) and AI technologies at the heart of this analytics opens up a treasure trove of operational and business insight. The question is, what to do with these insights? Here are five major use cases for speech analytics.
Only one to two percent of calls in an average contact center ever get listened to by either a supervisor or QA team member. Unless something bad happens, like complaints arising from cases of non-compliance, mis-selling or privacy breaches, those call recordings just sit on the shelf. While listening to them after the complaints have happened can help identify what went wrong, the damage has already been done.
Modern speech analytics software can rapidly analyse call recordings to spot compliance issues and either warn the agent, or escalate to a manager to take corrective action. Some solutions can even do this on-the-fly as calls are in progress. Knowing that every call is getting listened to and checked for compliance with regulatory rules, selling best practices, and privacy policies allows your business to avoid the reputational and financial damage that can arise from breaches in any of those areas.
Customers may be “king” but that doesn’t mean they always do the right thing. Speech analytics can not only identify but predict the likelihood of customer fraud.
A global general Insurance Company wanted to improve their current Fraud Risk prediction engine by utilising the call recordings between agents and customers. The objective was to determine the likelihood of ”fraud” during the claim lodgement call (motor vehicle) and enhance the fraud prediction rates to repudiate claims that did not warrant pay-outs.
Using the outputs of speech analytics (transcriptions, interaction analytics etc), a supervised machine learning module was built against a training data set (“Yes Fraud” and “No Non-Fraud” calls). After blind tests and ongoing improvements, the supervised learning model has been operationalised as a ‘customer voice’ data feed to the existing Fraud Risk engine. Customers are now triaged based on enhanced risk prediction.
The outcome has been very successful into the insurance company’s existing workflow. The uplift in the insurer’s risk engine is material and the reduction in the cost of claims is circa 1%. This is an ROI of over 1000%.
Most supervisors and QA staff only get to listen to a handful of calls per agent per week, if they’re lucky. It remains however an important tool for ensuring that agents are on the right track, and constantly improving by following their personal training and development roadmap.
The job of supervisors is to find proof of this improvement, or lack of it, by monitoring agent performance against set criteria over time. However when only a fraction of calls get listened to, the data is generally not conclusive either way – perhaps the supervisor just happened to pick a really bad call.
Speech analytics can listen to every call. By listening for keywords and phrases, ensuring the agent follows processes, and by analysing both agents’ and customers’ language and tone for sentiment, every call can be automatically scored against multiple criteria. What’s more, these rules will be applied perfectly consistently. As a result, you can improve overall performance with additional training to plug the skills and knowledge gaps identified. This also helps to improve the speed to competency for new recruits, and reduce churn in those all-important first few weeks and months.
Speech analytics can also help you to stay in touch with the emotional and mental well-being of your agents. Tone and sentiment analysis is used to assess both parties in an interaction. If an agent shows signs of stress, frustration, anger or other negative emotions, this will inevitably come through in the analysis. Managers can then be warned to intervene with a metaphorical arm around the shoulder, additional training, or even counselling.
It’s one thing to know and track your key Customer Experience metrics, like NPS (Net Promoter Score), Customer Satisfaction, Customer Effort, and so on, but quite another thing to understand why those scores are what they are.
Using speech analytics, you can take the measure of each customer’s tone of voice and words used to actually predict – quite accurately – the customer satisfaction or NPS rating the customer would have given for a particular call. This means that rather than relying on the few customers you do follow up with to arrive at those scores, you can effectively get a predicted satisfaction rating for every interaction.
Linguistic analysis can also pick out customers’ verbatim comments made to agents to help you understand exactly what they do and do not like about your products, service levels, and processes. Feed this information into your analysis of your metrics when you collect them to understand why customers give the scores they do, and what they would like you to do to improve them.
Even if you are already recording all calls and listening in to a small percentage of them for compliance and training purposes, that still means that most of the useful information in those interactions is not being taken on board.
While talking to your agents, your customers give up lots of incredibly detailed information in off-the-cuff comments related to such things as how they use your products, how your prices compare, how accessible and efficient your after-sales care is, and most importantly how you stack up against your direct competition. This is gold dust insight that you would normally need to conduct thousands of surveys to uncover – and it’s right there already on a dusty server in your unlistened-to call recordings.
Your sales and marketing teams should be poring over this information and using it to inform their decisions on everything from product development to pricing, to market positioning and what sales channels to use.
It’s an old adage that every interaction with a customer or prospect is an opportunity to move the relationship forward, instil loyalty by demonstrating that you deserve it, or make a sale. When it comes to sales, it is certainly better for the customer relationship to not be selling all the time, however there are times when it is appropriate. What’s difficult is pinpointing exactly the right time.
If you understood the factors that drove people to buy at a certain time and not others, or via a certain channel, you could make that process repeatable. Rather than guess or intuit this information, it’s much better to let your customers tell you why they take the decisions they do. Speech analytics allows you to identify what sentiments in what situations lead to what behaviours and what the results usually are.
You can then redesign processes to eliminate behaviours that don’t lead to sales and retention, and encourage ones that do. Your scripts and workflows can be written to take advantage of the best moments during interactions to make an upsell or cross-sell offer, ensuring that closing rates are maximized. Outbound contact programmes can also be used to follow up with customers who had, or whom the analytics predict are likely to have, a negative experience with a product or service.
There are many more use cases of speech analytics that you can likely think of. And the good news is that if you don’t have a budget to buy the technology, you can deliver call recordings to a service provider, who will analyse them and charge you for speech analytics as a service, as and when you use it.
Untapping the compliance, operational, sales, marketing, and general business insight hitherto hidden away in your call recordings can be incredibly transformational. Voice of the Customer information and Voice of the Employee programs are important initiatives now for contact centers. Speech analytics can supercharge both of them with more data than you could ever gather manually.